MICKAI Labs / Trust Agent

Semantic search studio

Embed your own small collection of sentences. Find which ones are closest to a question.

How this experiment works

MiniLM produces normalised 384-dimensional sentence embeddings. The query ranks sentences by their dot product.

What it does not prove

Cosine similarity is not truth or relevance certification. MiniLM has a limited context and domain coverage.

Try this challenge

Find a paraphrase with no shared nouns. Compare this with the lexical vector mode in Memory studio.

Related Trust Agent courses ยท Sources and licences

Enable JavaScript to run the experiment on your device.